What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Frontmatter

id	13076
title	Harness Endurance Benchmark — measurement & comparison machinery (Subjects A/B + deterministic load + event-loop-lag runner)
state	Closed
labels	enhancementaitesting
assignees	neo-opus-vega
createdAt	Jun 13, 2026, 9:31 AM
updatedAt	Jun 13, 2026, 4:26 PM
githubUrl	https://github.com/neomjs/neo/issues/13076
author	neo-opus-vega
commentsCount	0
parentIssue	13032
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
blockedBy	[]
blocking	[]
closedAt	Jun 13, 2026, 4:26 PM

Harness Endurance Benchmark — measurement & comparison machinery (Subjects A/B + deterministic load + event-loop-lag runner)

Closed v13.1.0/archive-v13-1-0-chunk-2 enhancementaitesting

neo-opus-vega commented on Jun 13, 2026, 9:31 AM

Parent: #13032 · Epic: #13012

The reusable measurement & comparison machinery for the Harness Endurance Benchmark — the falsifier scaffold #13032 needs before it can render a calibrated verdict. Split out from #13032 so the machinery can land now while #13032 stays open for the load-calibration → conclusive verdict → published numbers it still tracks.

Scope (what this delivers)

examples/harnessEndurance/shared/LoadProfile.mjs — deterministic, seeded (mulberry32) load: zero-dep markdown-append stream + keystroke cadence, byte-identical per {seed, config} so both subjects + the runner consume the IDENTICAL sequence.
Subject A (examples/harnessEndurance/neo/) — the Neo transcript surface (off-thread parse/diff via Neo.component.markdown.Component), self-driving the append stream.
Subject B (examples/harnessEndurance/comparator/) — an honest single-main-thread vanilla comparator (full re-parse each tick), at security parity (hostile input rendered inert).
Runner (test/playwright/e2e/HarnessEnduranceBenchmark.spec.mjs) — drives both subjects under the same load + samples main-thread event-loop lag (per-tick re-anchored; NOT rAF, which free-runs without vsync in headless Chromium) + heap, logging the cross-subject delta.

Acceptance Criteria

Deterministic load generator with unit-proven reproducibility.
Both subject apps boot + stream the load, manually runnable standalone.
One-command e2e runner samples a credible main-thread-health metric for both subjects + logs the delta.
Metric-credibility documented (event-loop-lag rationale; rAF-headless + accumulator-drift traps caught + fixed).

Out of scope (remains on #13032)

Load-calibration → the conclusive cross-subject verdict (no p-hacking-to-Neo-win).
Published numbers + methodology on public surfaces (negative result publishes with equal prominence).
keystroke→echo / frame-time / task-queue-depth metric layers.

Delivered by PR #13066.